The Integral $\int_{0}^{\infty}\frac{\text{sin}\ x}{x}dx$

Author(s): G. H. Hardy
Source: The Mathematical Gazette, Vol. 5, No. 80 (Jun. - Jul., 1909), pp. 98-103
Published by: The Mathematical Association
Stable URL: https://www.jstor.org/stable/3602798
Accessed: 13-05-2025 03:46 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

The Mathematical Association is collaborating with JSTOR to digitize, preserve and extend
access to The Mathematical Gazette

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms
98 THE MATHEMATICAL GAZETTE.
(3) When 0 is the median centre a=b=c=l1,
and x+y+z=O, is the line at infinity (IV. 4).
Hence A'B' jj AB, etc.
VI. Illustrative example.
Mentioned but not proved by Hamilton.
Unessential to the later sections.
1x2 - 21yz=O (a) represents a conic touching AB (z=O0) where (x- y)2=0,
i.e. at C', that is, (a) is the inscribed conic touching at A', B', C'.
Yyz=O (b) represents a (reciprocal, IX. (3)) circumscribed conic, for if
x=O, then y or z=O.
If P1, P2 be the poles of p-lx+my+nz=O w.r. to (a) and (b), we have
2(X1 - Y1 - )x=O are p;
4( Y2 + Z2)x=O r
SX -- F1- Z

.... ...........-............i

Y + rn

l+m-n.... .
But 2(l+m)-(+m - n)= 1+m+n.
Hence, by IV. (1) (a), OPxP2 are collinear.

VII. The meet line.

(1) The tangents at A, B, C, to a circumconic, meet the opposite sides
on "the meet line."
The general circumconic is yz+mzx +nxy= 0.
Tangent at x, y, z is Y(mZ+n Y)x= O.
Hence tangent at (1, 0, 0) is
ny +mz=-0
which cuts BC where x= O.
Hence meet line is Y+Imn
z = 0.
(2) For the circumconic lyz=O0.
Meet line is A" B" C". F. J. O. CODDINGTON.
(To be continued.)

THE INTEGRAL sin dx.

Jo x
No one can possibly welcome more cordially than I do the widening of the
school curriculum in mathematics that has taken place during the last
twenty years. Even I can remember the days when 'Conics' and
' Mechanics' were the privilege of a select few, and only a prodigy was
initiated into the mysteries of the Differential (much less the Integral)
Calculus. All this is changed now, and it is fairly safe (at least that is my
experience) to assume that a boy who has won a scholarship has learnt
something about the Integral Calculus at school.

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms
 THE INTEGRAL sindx. 99
I do not believe for a moment that the curriculum has been made harder.
These happy results are not due to any striking development in the school-
boy himself, either in intelligence, or in capacity or willingness for work.
They are merely the result of improvement in the methods of teaching and
a wider range of knowledge in the teacher, which enables him to discriminate,
in a way his predecessor could not, between what is really difficult and what
is not. No one now supposes that it is easy to resolve sin x into factors and
hard to differentiate it !-and even in my time it was the former of these
two problems to which one was set first.
Still there are difficulties, even in the Integral Calculus; it is not every
schoolboy who can do right everything that Todhunter and Williamson did
wrong; and, when I find my pupils handling definite integrals with the
assurance of Dr. Hobson or Mr. Bromwich, I must confess I sometimes
wonder whether even the school curriculum cannot be made too wide.
Seriously, I do not think that schoolboys ought to be taught anything about
definite integrals. They really are difficult, almost as difficult as the factors
of siti x: and it is to illustrate this point that I have written this note.

(1) x sin =
I have taken the particular integral

as my text for two reasons. One is that it is about as simple an example as

one can find of the definite integral proper-that is to say the integral whose
value can be expressed in finite terms, although the indefinite integral of
the subject of integration cannot be so determined. The other is that two
interesting notes by Mr. Berry and Prof. Nanson have been published
recently,* in which they discuss a considerable number of different ways
of evaluating this particular integral. I propose now to apply a system of
marking to the different proofs of the equation (1) to which they allude,
with a view to rendering our estimate of their relative difficulty more
precise.
Practically all methods of evaluating any definite integral depend ultimately
upon the inversion of two or more operations of procedure to a limitt-e.g.
upon the integration of an infinite series or a differentiation under the
integral sign: that is to say they involve 'double-limit problems.'$ The
only exceptions to this statement of which I am aware (apart, of course, from
cases in which the indefinite integral can be found) are certain cases in
which the integral can be calculated directly from its definition as a sum,
as can the integrals

fjrlog sin 0 dO, f log(1- 2p cos 0+p2)dO.

Such inversions, then, constitute what we may call the standard difficulty
of the problem, and I shall base my system of marking primarily upon them.
For every such inversion involved in the proof I shall assign 10 marks. Be-
sides these marks I shall add, in a more capricious way, marks for artificiality,
complexity, etc. The proof obtaining least marks is to be regarded as the
simplest and best.

* A. Berry, Messenger of Mathematics, vol. 37, p. 61 ; E. J. Nanson, ibid. p. 113.

t Whether this is true of a proof by 'contour integration' depends logically on what
proof of Cauchy's Theorem has been used (see my remarks under 3 below).
$For an explanation of the general nature of a 'double-limit problem' see my Course
of Pure Mathematics (App. 2, p. 420); for examples of the working out of such
problems see Mr. Bromwich's Infinite Series (passim, but especially App. 3, p. 414
et seq.).

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms
100 THE MATHEMATICAL GAZETTE.
1. Mr. Berry's8first proof.* This is that expressed by the series of equations

0sill x dxj= rnlim (e-" ,sin X dx= linimj e-ax illX dx

o x o a-+O - 0/ a--- o .X
a--O - a---0. o o
= lim fe-axdxf costxdt= lim dt e-axcostxdx

1fadt ()
- li I adt = lim arctan 1
a-o-. o ao + t2=a ()0a=
This is a difficult proof, theoretically. Each of the tirst two lines involves,
on the face of it, an inversion of limits. In reality each involves more: for
integration, with an infinite limit, is in itself an operation of repeated pro-
cedure to a limit; and we ought really to write

lim = lim lim = lim lir = lim lirnfX = lirn ?f.

O a-~0 X--ao a--0 X--o a--0 o a--0 X-*a 0 a- --0
Thus two inversions are involved in the first line; and so also in the
second. This involves 40 marks. On the other hand this proof is, apart
from theoretical difficulties, simple and natural : I do not think it is
necessary to add more marks. Thus our total is 40.
2. AMr. Berry's second proof. This is expressed by the equations

0silldx
1 a)di 11 s+x
f(i+i)w sill IXxx
dx=-
fo--sXd2 - 2 x 25 1 ir X
0 f- 2 o -0 oa + i~r
S sin x cosec xdx = r.

It is obvious that this has a great advantage over the first proof in that
only one inversion is involved. On the other hand the uniform convergence
of the series to be integrated is, as Mr. Berry remarks, 'not quite obvious.'
Moreover the fact that the cosecant series contains the terms
11
which become infinite at the limits, although it does not really add to the
difficulty of the proof, does involve a slight amount of additional care in its
statement. I am inclined to assign 15 marks, therefore, instead of 10 only,
for this inversion. For the first three steps in the proof I assign 3 marks
each; for the use of the cosecant series 4. Thus the total mark may be
estimated at 9 + 15 + 4 = 28.
3. Mr. Berry/'s third proof (by contour integration). It is naturally a little
harder to estimate the difficulties of a proof which depends upon the theory
of analytic functions. It seems to me not unreasonable to assign 10 marks
for the use of Cauchy's theorem in its simplest form, when we remember
that the proof of the theorem which depends upon the formulat

J f ap d) = (pdx+qdy)
ff (a~ix D Y ) f"Y-J ~U~TY CY

*I mean, of course, the first proof mentioned by Mr. Berry-the proof itself is
classical.
t+Forsyth, Th eory of Functions, p. 23.

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms
 THE INTEGRAL sinx-dx. 101
involves the reduction of a double integral partly by integration with respect
to x and partly by integration with respect to y. The actual theoretical
difficulty of this proof of Cauchy's theorem I should be disposed to estimate
at about 20; but half that amount seems sufficient for the mere use of a
well known standard theorem.*
The particular problem of contour integration with which we are con-
cerned is, however, by no means a really simple one. The range of integration
is infinite, the pole occurs on the contour; we have to use the fact that

lim limr + x- dx=- dx.

R=0O e=O -R 4 - ) x
For this I assign 8 marks; for the evaluation of the limit of the integral
round the ' small semicircle' another 8. There remains the proof that the
integral round the 'large semicircle' tends to zero. As Mr. Berry points
out, the ordinary proof of this is really rather difficult; I estimate its
difficulty at 16. Mr. Berry has suggested a simplification of this part of
the argument; the difficulty of his proof I estimate at 8. Thus we have
10+8+8+16=42 marks for the ordinary proof, and 10+8+8+8=34 for
Mr. Berry's modification of it.
4. Prof. Nanson's proof. This proof, I must confess, does not appeal to
me. Starting from the integral
a cos cx
--=- dz,
0J a + 2

where a and c are positive, we show, by a repeated integration by parts, and

a repeated differentiation under the integral sign, that u satisfies the equation

dau= C2U.
da2

As the upper limit is infinite we ought to assign 40 marks for this,

following the principles we adopted in marking the first proof; but as the
two operations of differentiation involve inversions of the same character I
reduce this to 30, adding 6, however, for the repeated integration by parts.
The transformation x= ay shows that u is a function of ac, so that
u=Aea"+Be-1a, where A and B are independent of both a or c. This step
in the proof I assess at 4.
That A =0 is proved by making c tend to o and observing that
adx
For this I assign 6 marks.
That c=jrr is proved by making c=O. Here we assume the continuity of
the integral, and this should involve 20 marks. But as the proof is simple
I reduce this to 12. Thus by the time that we have proved that

q4 = ?i"M-ca)

we have incurred 30+6+4+6+12= 58 marks.

We have then
f asinmxdx adx f cos
Jx(a + X2 Jo a+x coscxd
= udc=t (1-e--).

* Goursat's proof is better and more general, and does not involve my inversion of
limit-operations, but its difficulties are far too delicate for beginners.

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms
102 THE MATHEMATICAL GAZETTE.
This involves 20 marks. We then obtain
i" X SiIl mx 1.,-am
J a2 +2 d
by two differentiations with respect to m. This should strictly involve 40,
and I cannot reduce the number to less than 30, as the second differentiation
is none too easy to justify. The result then follows by multiplying the
last formula but one by a and adding it to the second.
I do not think that the 58+20+30=108 marks which we have assigned
are more than the difficulties of the proof deserve. But of course it is hardly
fair to contrast this heavy mark with the 40, 28, 42, 36 which we have
obtained for the others; for Prof. Nanson has evaluated the integrals
f si nmx dx, cOS mx dxfx sin mxdx
.o (a+ x2) d.x ?o + J a2 + X2
as well as the integral (1).
5. Mr. Michell's proof.* This depends on the equations
if sin x d li fsin dx
- x h--0, H---oo h
- lim d x e-x in cos(x cos 8) dO
h--O, - H-~I h d
i= - lim. dx d cos(xc cos dx
i tdO e-xginO
h--*0, H--0oo fl, h

u= - in e- Bl cos(F0cos O)-e-(si cos(h cos O) dO

h-*O, H--- J 0

I have myself for several years used a proof, in teaching, which is in

principle substantially the same as the above, though slightly more simple
in details and arrangement, viz.:

sin x dx= dx ei(t+xeit) dt

Jo 0
_1 eitdtf eixeitdx

_1_ dt=-rr.

The successive steps of the proof can of course be stated in a form free
from i by merely taking the real part of the integrand. To justify the
inversion we have to observe that

er itdt f eixetdx= f dxfo eit+ixtdt

for any positive value of X, in virtue of the continuity of the subject of

integration, and that

eJdt eid etdx = I fexetdtI < .foe-x1ntdt

<2J 0e-x sill Idt< 2foJre-2Xtl/wdt

< 7(1e - e-x)X,

* This proof is also referred to by Prof. Nanson. The only proofs which are
'classical are 1, 2, 3, viz. the three discussed by Mr. Berry.

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms
 THE INTEGRAL vin . 103
the closing piece of argument being essentially the same as that at the end
of Prof. Nanson's paper. Hence
frf =lirmf =lir m ffr =
" --+m -0 0 " " X- 0-0
This proof involves an inversion of a repeated integral, one limit being
infinite. This involves 20 marks. To this I add 15 on account of the
artificiality of the initial transformation, deducting 3 on account of the ex-
treme shortness and elegance of the subsequent work. Thus we obtain
20+15-3=32 marks. To Mr. Michell's proof, as stated by Prof. Nanson, I
should assign a slightly higher mark, say 40.
We may therefore arrange the proofs according to marks, thus:
1. Mr. Berry's second proof, - - 28
2. Mr. Michell's proof (my form), - 32
3. Mr. Berry's third proof (his form), - - - 34
4. Mr. Berry's first proof, - - 40
5. Mr. Michell's proof (his form), - - - 40
6. Mr. Berry's third proof (ordinary form), - - 42
7. Prof. Nanson's proof, - 108
And this fairly represents my opinion of their respective merits : possibly,*
however, I have penalized Nos. 2 and 5 too little on the score of artificiality
and 4 too much on the score of theoretical difficulty. I conclude, however,
with some confidence that Mr. Berry's second proof is distinctly the best.
But whether this be so or not, one thing at any rate should be clear from
this discussion: whatever method be chosen, the evaluation of the integral (1)
is a problem of very considerable difficulty. This integral (and all the other

mathematics. G. H. HARDY.
standard definite integrals) lie quite outside the legitimate range of school

REVIEWS.
The Foundations of Mathematics: A Contribution to the Philosophy
of Geometry. By DR. PAUL CARUS. Chicago: The Open Court Publishing
Company. 1908. Pp. 141.
This book is a more or less popular exposition of a philosophy of geometry
which is, in its main outlines, derived from Kant. The main title, if uncorrected
by the sub-title, would be somewhat misleading, since the foundations of
arithmetic and analysis are not discussed, but only the foundations of geometry.
The author begins by a brief account of the development of non-Euclidean
Geometry, which is followed by much longer chapters "on the philosophical
basis of mathematics" and on "mathematics and metageometry." The historical
chapter, though it does not profess to give more than a sketch, might with advan-

*This I know to be Mr. Berry's opinion. He would put 2 below 3 and 4, and 5
below 6. To compare 3 and 4 is very difficult, and they might fairly be bracketed.
Since writing this note I have recollected another proof which I had forgotten, and
which is given in ? 173 (Ex. 1) of Mr. Bromwich's Infinite Series. Mr. Bromwich there
proves that, provided the real parts of a and b are positive or zero,

Putting a= 1, b= i, we obtain

fO(e-x-cos x) d =0, sin fxdx=jr.

This proof, as presented by Mr. Bromwich, should be marked at about 45. It has
the advantage of giving the values of several other interesting integrals as well, so that
(as in the case of Prof. Nanson's proof) it is hardly fair to contrast this mark with the
considerably lower marks obtained by some of the other proofs.

This content downloaded from 128.103.147.149 on Tue, 13 May 2025 03:46:53 UTC
All use subject to https://about.jstor.org/terms